A1087
Title: New procedure for controlling false discovery rate in Cox model
Authors: Hengjian Cui - Capital Normal University,Beijing,China (China) [presenting]
Abstract: A novel feature selection method for the Cox model in high-dimensional data analysis is developed. The method is constructed under the framework of the FDR control for multiple testing, and the multiple data-splitting strategy is adopted. For each splitting, the data is divided into two disjoint parts. The first part of the data is used for feature selection, and multiple tests are conducted for the set of selected features in the two parts. Then, the z-values of the statistics are aggregated to control the FDR, and the set of important features is chosen by rejecting the null hypotheses. The asymptotic theory of FDR control for the proposed method is established under mild conditions. The finite sample performance of the feature selection procedure is evaluated by Monte Carlo simulations. It is shown that the proposed new procedure effectively controls the empirical FDR. The new approach is illustrated through a real dataset from the diffuse large-B-cell lymphoma study.
